Global patterns in leaf ¹³C discrimination and implications for studies of past and future climate

Abstract

Fractionation of carbon isotopes by plants during CO₂ uptake and fixation (Δ_leaf) varies with environmental conditions, but quantitative patterns of Δ_leaf across environmental gradients at the global scale are lacking. This impedes interpretation of variability in ancient terrestrial organic matter, which encodes climatic and ecological signals. To address this problem, we converted 3,310 published leaf δ¹³C values into mean Δ_leaf values for 334 woody plant species at 105 locations (yielding 570 species-site combinations) representing a wide range of environmental conditions. Our analyses reveal a strong positive correlation between Δ_leaf and mean annual precipitation (MAP; R² = 0.55), mirroring global trends in gross primary production and indicating stomatal constraints on leaf gas-exchange, mediated by water supply, are the dominant control of Δ_leaf at large spatial scales. Independent of MAP, we show a lesser, negative effect of altitude on Δ_leaf and minor effects of temperature and latitude. After accounting for these factors, mean Δ_leaf of evergreen gymnosperms is lower (by 1–2.7‰) than for other woody plant functional types (PFT), likely due to greater leaf-level water-use efficiency. Together, environmental and PFT effects contribute to differences in mean Δ_leaf of up to 6‰ between biomes. Coupling geologic indicators of ancient precipitation and PFT (or biome) with modern Δ_leaf patterns has potential to yield more robust reconstructions of atmospheric δ¹³C values, leading to better constraints on past greenhouse-gas perturbations. Accordingly, we estimate a 4.6‰ decline in the δ¹³C of atmospheric CO₂ at the onset of the Paleocene-Eocene Thermal Maximum, an abrupt global warming event ∼55.8 Ma.

Human perturbation of the global carbon (C) cycle is potentially far greater in rate and magnitude than variations in the recent past, pushing predictions of future climate beyond the calibration range of models based on modern and near-modern observations. Robust predictions of future impacts of rising CO₂ require not only extrapolation of ecological patterns along modern environmental gradients but also insights gained from changing ecological patterns at times of high CO₂ and hot climate in the geologic past (1 and 2). Global patterns of variation in leaf carbon isotope () values potentially record climate-driven changes in modern plant physiology and biogeochemistry. An understanding of factors controlling plant fractionation (Δ_leaf) at the global scale will improve interpretations of past changes in climate and ecology recorded in ancient terrestrial sedimentary organic carbon (2). Patterns in of living plants at the global scale, however, are unresolved in spite of abundant published data at smaller spatial scales.

In living plants, values reflect the balance of photosynthesis and stomatal conductance and their coupled response to the environment (3). Edaphic factors (e.g., water availability, altitude, temperature) and plant attributes (e.g. phylogeny and leaf traits) can influence values (4–6). The relative importance of these factors at the global scale is not known, nor is it clear how they might drive the variations in ancient values recorded in either terrestrial organic carbon () or δ¹³C values of plant biomarkers.

During extreme climate events in the past, changes in the isotope ratio of atmospheric CO₂ () reflect perturbations in atmospheric C fluxes, therefore accurate estimates of are central to estimating past climate sensitivity to changes in pCO₂ (7). Values of from sedimentary rocks are widely used to estimate of ancient plants and infer changes in (8). However, the offset between and (i.e., Δ_leaf) varies with climate and plant characteristics and such variations must be accounted for when estimating atmospheric C isotope excursions (CIE) from ancient plant-derived C. For example, during the Paleocene-Eocene Thermal Maximum (PETM), a period of rapid global warming 55.8 million years ago (9), the CIEs in terrestrial organic carbon (TOC) and atmospheric CO₂ likely differ because of changes in Δ_leaf that accompanied plant community shifts, warming (∼5 °C), decreasing precipitation and increases in atmospheric CO₂ (2).

In this study, we provide predictive relationships for Δ_leaf variability of woody C₃ plants at the global scale from analysis of published values, plant functional types, biome, climate, and geography. We use these relationships to show how environmental and ecologic change during the PETM influenced plant fractionation and therefore the CIE recorded by fossil leaf waxes.

Results and Discussion

Leaf Carbon Isotope Fractionation.

We used Δ_leaf in our analyses as it controls for variation in (10). Δ_leaf in C₃ plants is a function of the fractionation associated with CO₂ diffusion (4.4%) and photosynthetic fractionation by Rubisco (27%). For C₃ plants, the concentration of CO₂ in the substomatal cavities of leaves (c_i) is directly proportional to Δ_leaf when the atmospheric CO₂ concentration (c_a) is held constant (10) (but see ref. 11). In turn, c_i depends on the flux of CO₂ into the leaf, which is largely regulated by stomatal conductance (g_s) and the flux of CO₂ removed from the leaf for C fixation by assimilation (A) (10). Values of Δ_leaf generally decrease with reductions in water availability, reflecting a down-regulation of g_s and increased water-use efficiency (WUE) (10, 12). Thus, Δ_leaf is a time-integrated measure of WUE, although leaf temperature (T), mesophyll conductance and differences in c_a are potential confounding factors (11 and 13). Confounding effects are reduced, however, because mesophyll conductance is correlated with A (and with g_s) (11 and 14) and trees appear to maintain leaf T within a narrow range (21.4 ± 2.2 °C) across a wide range of ambient T (15). A and g_s also vary with plant phylogeny, growth form and leaf traits, thus Δ_leaf and its response to the environment may differ for plant functional types (PFTs) defined by these attributes (4).

Data.

Our dataset, provided in Dataset S1, includes 334 species from 75 woody plant families (including trees and shrubs) present at 105 geographic sites distributed across 5 continents (representing 3,310 measurements of individual plants). The result is 570 unique species-site combinations represented by a mean Δ_leaf value calculated from multiple individuals of each species at a site. These sites represent eight biomes with mean annual temperature ranging from -10° to 28 °C and mean annual precipitation (MAP) from 147 to 3,700 mm per year, capturing the broad range in which higher plants occur (SI Appendix).

Precipitation.

MAP is positively correlated to global Δ_leaf values (p < 0.0001, R² = 0.55; Fig. 1) and is the strongest predictor of Δ_leaf among the environmental parameters in our dataset. To our knowledge this study represents the largest spatial scale at which this correlation has been documented, though it has been seen in some, but not all, regional studies and with considerably lower explanatory power (16–20). The strong correlation of MAP with Δ_leaf suggests that as water availability increases, stomatal limitations on c_i and A relax and WUE declines (i.e., g_s increases relative to A), leading to greater absolute C fixation (3 and 10). These apparent stomatal limitations on leaf-level A likely scale up to the ecosystem level and help explain global variation in gross and net primary production, which also varies with MAP (21).

Fig. 1.

Effect of mean annual precipitation (MAP) on Δ_leaf (n = 506). MAP accounts for 55% (p < 0.0001) of the variability in Δ_leaf in a linear regression model. Points are coded by biome: tropical rain forest (TRF), evergreen warm mixed forest (EWMF), tropical seasonal forest (TSF), cool-cold deciduous forest (CCDF), cool-cold evergreen forest (CCEF), cool-cold mixed forest (CCMF), tropical deciduous forest (TDF), xeric woodland scrubland (XWS). In a regression model using site-means (Δ_leaf averaged by geographic location), MAP accounts for 52% (p < 0.0001) of the variability in Δ_leaf (SI Appendix).

The relationship between MAP and Δ_leaf is consistently strong within the continents of North America (NA) and Asia (SI Appendix), but not Europe (EU; R² = 0.025, p = 0.125). Extensive analysis of European data does not reveal any single environmental variable (or simple combination of variables) that accounts for more than 20% of the variation in Δ_leaf across all plant types. The lack of strong MAP effects on Δ_leaf in EU (see also ref. 16) and elsewhere (17–19) may arise from heterogeneity in growing season water availability that is not captured by MAP, due to differences in the seasonality of rainfall across sites (i.e. EU includes areas with wet winters in the south and wet summers in the north) or the presence of sites where soil water does not come from local precipitation. Better measures of water availability during growth may yield different results. For example, in our NA dataset, summer precipitation (June to August) is a better estimate of water availability than MAP for deciduous angiosperms and is more strongly correlated with Δ_leaf (R² = 0.83 and 0.57 respectively, p < 0.0001, n = 36 species-site combinations, excluding tropical sites and one California site with a wet winter).

Altitude and Other Factors.

We assessed the influence of other factors on Δ_leaf using bivariate partial regression models that account for the covariance of MAP with other predictors (SI Appendix) and its correlation with Δ_leaf. This approach is justified by our observation that MAP emerges as the strongest single predictor of Δ_leaf and because the influence of MAP on Δ_leaf is supported by theory and other observations. The partial regression model with altitude as a secondary predictor is the only model with notable explanatory power (R² = 0.13, p < 0.0001) indicating that altitude has a weak, negative influence on Δ_leaf that is statistically independent of its covariance with MAP. The combination of MAP and altitude explains 61% of the global variability in Δ_leaf in a multiple regression model (p < 0.0001), and regression models for data-rich continents show similar results (SI Appendix). Elevation effects on Δ_leaf have been observed previously (3 and 20), although altitude is rarely evaluated independently of MAP. The responsible mechanisms remain uncertain (3) and could be related to T, vapor pressure, oxygen partial pressure, irradiance, or other factors.

MAP and altitude effects on Δ_leaf are evident when regressions are performed on data grouped by PFT (SI Appendix), indicating that their influence is independent of community changes along elevation or precipitation gradients. The effects of MAP and altitude are still apparent when regressions are performed separately for well represented genera in our database (SI Appendix). Phylogenetic variation is reduced within genera, making it unlikely that the global correlations of Δ_leaf with MAP (Fig. 1) and altitude are driven exclusively by differences in the taxonomic composition of floras in different regions. Rather, we suggest Δ_leaf is evolutionarily and physiologically plastic in response to changes in water availability.

Unexpectedly, other factors (e.g. T, latitude, evapotranspiration) were not significantly related to Δ_leaf after controlling for MAP and altitude in regression models of the entire dataset. Negligible effects of these factors could indicate that A and g_s covary more tightly with T and other factors than with changes in MAP and altitude. Covariance of T with MAP, altitude, or PFT distribution may also mask its influence (SI Appendix, see below).

Plant Functional Types.

Despite the considerable variance in Δ_leaf explained by MAP and altitude, much remains unexplained, as evident by the ∼8% range in Δ_leaf at any given MAP value (Fig 1; Dataset S1). Microclimate, ecosystem structure, and differences in plant traits among the species and PFTs must explain some of this variability. Several studies have reported localized differences in Δ_leaf among woody PFTs and similar results have been observed in metaanalyses of wood δ¹³C (6). Yet previous metaanalyses of Δ_leaf of C₃ plants have not found evidence of a PFT effect (22), with the exception that grasses differ from woody plants (4). This likely reflects inadequate control for environmental influences on Δ_leaf that are independent of PFT. On the other hand, accounting for the uneven distribution of PFTs along environmental gradients may reveal additional environmental controls on Δ_leaf. When regression models are analyzed separately for each PFT, small latitude and T effects on Δ_leaf are evident for deciduous angiosperms (DA) and evergreen gymnosperms (EG), the PFTs that span the greatest range in latitude and T (SI Appendix). For DAs and EGs, increases in latitude or decreases in T result in slightly higher Δ_leaf than predictions based on MAP and altitude alone, indicating limitations on carbon assimilation due to low T or irradiance.

We used a multiple regression model with MAP and altitude as predictors to capture the dominant environmental influences on Δ_leaf, and explored whether PFTs explain any residual variation in Δ_leaf. This approach reveals significant differences in Δ_leaf among PFTs (SI Appendix), with EGs between 1.0% and 1.5% lower than the others. More strict control of confounding environmental influences on Δ_leaf is achieved by limiting PFT comparisons to plants at the same geographic site (Fig. 2; SI Appendix). With this constraint, DAs and evergreen angiosperms (EA) have higher Δ_leaf than EGs, by 2.7% and 2.2% respectively. PFT effects on Δ_leaf are not solely related to phylogeny or leaf habit. These differences may be related to genetic or phenotypic differences in A and g_s mediated by leaf morphology, hydraulic architecture, rooting depth, leaf T, and/or mesophyll conductance (3, 10, 11, 13). These traits also vary among species within the same PFT; future studies should examine this variation.

Fig. 2.

Differences in Δ_leaf between cooccurring plant functional types (PFTs). For each geographic site, mean Δ_leaf values of each species were averaged to produce a mean Δ_leaf for each PFT. Box and whisker plots show the median, upper and lower quartiles, and maximum and minimum values, with outlier values shown as black dots. Number of sites per PFT comparison are as follows (deciduous angiosperm, DA; deciduous gymnosperm, DG; evergreen angiosperm, EA; evergreen gymnosperm, DG): DA - EA = 16, DA - EG = 17, DG - EG = 21, EA - EG = 12. DA-DG comparison is shown to highlight differences between Larix (triangle) and Taxodium (star).

Deciduous and evergreen species have contrasting nutrient use and retention strategies (23) that may underlie their dominance in different ecosystems. Our results support the notion that WUE differences are also an important component of plant strategies related to resource availability and likely help explain PFT dominance patterns. Lower Δ_leaf implies greater WUE of EGs relative to other woody plants, conferring a potential advantage where water is limiting and a limitation on A when water availability is high (24).

Biome Patterns.

Mean Δ_leaf values calculated by biome type (Fig. 3; SI Appendix) differed significantly, with greatest fractionation in tropical rain forests (mean Δ_leaf = 23.4%) and lowest in xeric woodland/scrublands (mean Δ_leaf = 17.3%), which compare well with biome level Δ_leaf patterns predicted by Kaplan et al. (5). We observe a similar pattern when mean Δ_leaf for each biome is calculated separately for angiosperms and gymnosperms (SI Appendix). Biome type alone explains more Δ_leaf variation than MAP (R² of 66% and 55%, respectively, SI Appendix), because it captures other factors related to Δ_leaf, including the spatial distribution of PFTs. PFT influences at the biome scale are readily apparent; cool-cold deciduous forests have higher Δ_leaf values (by 1.2%) than cool-cold evergreen forests (Fig. 3; see also Kaplan et al. (5)). The DA-EG differences in Δ_leaf (1.0% to 2.7%) calculated for each biome are consistent with results based on residuals of multiple regression (1.0–1.5%; SI Appendix) and the paired-site comparison (2.7%; Fig. 2).

Fig. 3.

Box and whisker plots of Δ_leaf values by biome. Biome accounts for 66% (p < 0.0001) of the variability in Δ_leaf in an ANOVA model. Number of samples per biome are as follows (see Fig. 1 caption for abbreviations): TRF = 206; EWMF = 29; TSF = 47; CCDF = 5; MED = 5; CCEF = 53; CCMF = 78; TDF = 26; XWS = 59. Statistical tests of means are shown in SI Appendix.

Implications for Models.

Kaplan et al. (5) recently extended a coupled vegetation-biogeochemical model (BIOME4) to predict Δ_leaf at the biome and PFT scale. Our database for C₃ woody plants provides a means for full evaluation of BIOME4’s ability to reproduce global patterns in Δ_leaf. Since Δ_leaf records the time-integrated balance of A and g_s, global patterns in Δ_leaf can be used to assess the ability of physiological models to predict the balance between leaf C and H₂O fluxes. Further, incorporating Δ_leaf into coupled vegetation-biogeochemistry models have potential to enable predictive mapping of spatial variability in and , which could improve studies of the effect of climate on modern and ancient organic matter.

Geologic Implications.

The importance of water in regulating Δ_leaf values is often invoked in environmental and atmospheric reconstructions from (1). More recently, the role of PFTs in controlling has been identified (2 and 25). Our results show that variation in PFT and MAP independently result in differences in Δ_leaf of up to several % (Fig. 1 and 2), and that biome may be a powerful integrator of PFT, rainfall, and other effects on Δ_leaf and (Fig. 3). These quantitative relationships can be used to understand temporal and spatial variations in provided that MAP and PFT (or biome) can be estimated in the geologic past.

There are uniformitarian assumptions in using Δ_leaf patterns among extant plants to interpret the vegetation of the geological past, however two factors encourage us to proceed. First, the morphological and physiological responses of C₃ plants to water stress are constrained by their fundamental anatomy and biochemistry (26), making it likely that the relationship of water availability and Δ_leaf has been similar in direction and magnitude over geological time. MAP and altitude effects on Δ_leaf are apparent within as well as among genera, implying that independent lineages respond similarly and quickly. Second, a recent study by Crisp et al. (27) shows that plant lineages rarely shift biomes during evolution, suggesting niche conservatism and supporting the careful use of biome type to constrain Δ_leaf values. Nonetheless, applying modern correlations of Δ_leaf with environment to the distant geological past may be problematic because of differences in major plant groups prior to the origin of angiosperms in the Cretaceous.

Pollen and leaf fossils document spatial and temporal changes in PFT as well as T and precipitation (9 and 28). Plant biomarkers (chemical fossils), such as n-alkanes (from leaf waxes) or terpenoids (defense compounds) are also preserved geologically (9) and have fewer diagenetic or source effects on their isotopic composition than (29). Plant waxes in ancient sediments reflect productivity-weighted inputs from all PFTs contributing to that deposit. The resulting mixed molecular signal can be used to estimate a value that integrates plant biomass within the catchment. In contrast, terpenoids provide PFT specificity because tricyclic diterpenoids are unique to woody gymnosperms and pentacyclic triterpenoids are unique to woody angiosperms. Therefore, ratios of these terpenoids, after accounting for production and preservational biases, may provide estimates of PFT (30). Also, terpenoid δ¹³C values can be used to calculate PFT-specific Δ_leaf values (25).

Heterogeneity in rainfall and PFT dominance across ancient landscapes presumably created spatial patterns in Δ_leaf just as they do now, even at small spatial scales. To understand differences in , and the magnitude of CIEs recorded in different regions at the same time, and improve chemostratigraphic correlations using , we should account for biome (5), precipitation (9), and PFT (2 and 25) effects on Δ_leaf values using relationships like those reported above.

An important goal in studies that document temporal variation in is to reconstruct the sources and fluxes of C to the ancient atmosphere (7). and have been used to directly represent (4 and 31), but this approach is fraught with uncertainty (32). It has been suggested that using a simple numerical offset to convert to under nonlimiting water conditions is sufficient to control for environmental impacts (4). To the contrary, we argue that secular changes in PFT and climate must also result in significant—but often unaccounted—effects on Δ_leaf. Previous studies averaged multiple values to remove variation caused by environmental factors (4, 31, 32), ignoring insights about paleoecology and climate that might be inferred from variations.

We use an example from the PETM to illustrate how our study of modern Δ_leaf can improve interpretation of ancient and (Fig. 4; see SI Appendix). Smith et al. (2) documented a shift from a mixed angiosperm-conifer flora (∼25% angiosperm) in the late Paleocene to an angiosperm flora (∼100%) at the onset of the CIE, and inferred that the 5% decrease in (estimated from n-alkanes) reflected both a decrease in (3–4%) and an increase in Δ_leaf caused by a shift from conifers to angiosperms (1–2%, (2, 9)). Although the dominant late Paleocene conifers in the Bighorn Basin (Metasequoia and Glyptostrobus (9 and 33)) were deciduous, they probably had low Δ_leaf, as do their extant relatives (34 and 35) (Fig. 2). Yet the PETM in Wyoming is also marked by a change from wetter conditions (MAP ∼ 1,380 mm/yr) in the late Paleocene (28) to drier conditions (MAP ∼ 800 mm/yr) at the onset of the event (9), which should have decreased Δ_leaf. When both PFT and MAP effects on Δ_leaf are taken into account (Fig. 4; see SI Appendix), the decrease in Δ_leaf from declining MAP is countered by the increase in angiosperms, resulting in a net Δ_leaf increase of 0.2% and an estimated negative CIE in of 4.6%. This estimate of the atmospheric CIE is larger than that calculated by Smith et al. (2), who argued the CIE in was higher than the atmospheric CIE because of the loss of conifers, and that marine sediments recorded the true CIE. Our Δ_leaf correction for both PFT and MAP effects on suggests the relatively large CIE recorded in the Bighorn Basin and other terrestrial sites (36) may equal the atmospheric CIE. This larger CIE falls within the range recently reported from well-preserved shallow marine carbonate records (-3.5% to -5.0%) and supports the notion that -3.5% is too small (37). A larger atmospheric CIE implies the carbon released at the onset of the PETM was more depleted or greater in mass than presently thought. Adjusting other terrestrial CIE records for biome, MAP, and PFT effects should further refine and reconcile estimates of the magnitude of the CIE.

Fig. 4.

Conceptual diagram of reconstruction for latest Paleocene to earliest Eocene sediments in the Bighorn Basin (WY). The figures are constructed as follows: (A) values as inferred from δ¹³C values of n-C₃₁ alkanes (2), (B) percent of fossil leaves that are angiosperms, (C) mean annual precipitation (MAP) derived from fossil leaf metrics (9, 28), and (D) modeled net Δ_leaf from our Δ_leaf-MAP expressions for evergreen gymnosperm (EG; triangles) and deciduous angiosperm (DA; squares) PFTs (see SI Appendix). Each model was scaled by % angiosperm to derive a net Δ_leaf (dashed line). is denoted with a dashed line on A and is derived from the net Δ_leaf and values. The negative CIE in at the base of the Paleocene-Eocene Thermal Maximum (PETM) is 4.6% (difference between latest Paleocene value and average of initial PETM values). See SI Appendix for discussion of calculations and suggested use of these models.

Relationships based on modern plants may not fully quantify the response of Δ_leaf to greenhouse climate conditions (high T and pCO₂) such as those of the PETM (12, 13, 38). Given the multiple functions of stomata (water loss prevention, CO₂ provision for A, evaporative cooling), it is unlikely that plants could simultaneously maintain homeostasis with respect to leaf T (15) and c_i or Δ_leaf (12) during periods of rapid global change. Studies on Δ_leaf of extant relatives of ancient plants under high temperature and pCO₂ conditions and varying degrees of water stress are needed to better interpret changes in and during the PETM and other events.

Conclusions.

We document global patterns in Δ_leaf among woody plants, including: (i) a strong positive correlation with MAP (R² = 0.55), (ii) Δ_leaf values 1% to 2.7% lower for EGs than other woody PFTs, and (iii) differences up to 6% in Δ_leaf among biomes (R² = 0.66). The relationship of Δ_leaf with MAP shows that the balance of leaf gas-exchange in modern plants is largely driven by water availability, consistent with the role of water in driving global trends in ecosystem-scale primary productivity. By revealing global-scale relationships of MAP, PFT, and biome with Δ_leaf, our results uniquely enable applications of these patterns for understanding global-scale processes, such as extrapolation of the distribution of PFTs and leaf carbon and water fluxes under future climate scenarios. In geologic studies, our models of Δ_leaf provide tools for interpreting spatial and temporal variation in and , leading to enhanced understanding of paleoecology and atmospheric CO₂ during climate events that serve as analogs of the near future. Our results warn against the assumption in some geologic studies that Δ_leaf is invariant in space and time and emphasize that estimating from or requires information about biome, PFT, and paleoclimate. We use this approach to produce a refined estimate of the atmospheric CIE (-4.6%) during the PETM and suggest that it be extended to other PETM CIE records to better estimate T sensitivity to greenhouse gases. Such work will aid our understanding of plant response to extreme climate change both in the geologic past and in the coming century.

Methods

We extracted values for woody trees and shrubs from 45 publications and one new study (SI Appendix). We excluded values from juvenile, fertilized, or watered plants, immature or shaded leaves, and understory shrubs. All values were converted to using the Farquhar et al. (10) equation () and values estimated for the year of sampling (39) unless reported. Estimating may cause bias in Δ_leaf due to seasonal and latitudinal gradients in . However, this gradient is < 0.5% during the growing season (40). Species means were calculated for each geographic site to remove within-species variability. For a subset of sites where values from multiple PFTs were reported (n = 53), we calculated paired PFT differences at each site by averaging all species in each PFT and then calculating differences between PFT. For each geographic site, environmental factors were extracted from the publication or derived from global databases (SI Appendix) and biome classifications were assigned based on the BIOME4 model, descriptions in the original publication, and environmental factors. We analyzed the data using linear, least squares regression and ANOVA (SAS JMP 7.0). Δ_leaf values and latitude are approximately normally distributed. MAP was log ₁₀-normally distributed and transformed accordingly. Altitude was approximately normally distributed after square root transformation. We performed pairwise comparisons of means using the Tukey-Kramer HSD. Statistics reported within the text and appendices were performed at the site-species combination level, unless otherwise noted (n = 570 or less, depending on the model). Similar results were obtained from statistical models at the site level (SI Appendix) using means for all species at a site (n = 73 or less).